Distinction of gases with a semiconductor sensor depending on the scanning profile of a cyclic temperature

Abstract

A gas-sensing system based on a dynamic nonlinear response is reported to improve the selectivity in the sensor response toward sample gases. A cyclic temperature composed of fundamental and second harmonics was applied to a SnO₂ semiconductor gas sensor and the resulting conductance of the sensor was analyzed by fast Fourier transformation (FFT). The dynamic nonlinear responses to the gas species were further characterized depending on the scanning profile of the temperature. These characteristic sensor responses under the application of second-harmonic perturbation were theoretically considered based on a reaction–diffusion model for the semiconductor surface.

---

1. Introduction

An important consideration in the development of chemical sensors is that they should not only be able to rapidly detect a target chemical species but also that they should exhibit high chemical selectivity by eliminating interference even for an actual sample composed of mixtures or under variable environments. In general, the information obtained from semiconductor sensors is one-dimensional for each detector: the concentration of a certain gas species can be determined by the “d.c. change” in the resistance (or conductance) of a semiconductor gas sensor, and information on the “temporal axis” is ignored.^1–3 However, most semiconductor gas sensors are not selective enough to detect a single chemical species due to the principle of detection. Thus, the sensor responds to the most reducing gases, such as hydrocarbon and alcohol, and therefore it is impossible to oxidize only a specific gas in a gaseous mixture for detection.

Therefore, one-dimensional information is inadequate for distinguishing between the responses to the sample and other interfering gases. To overcome this problem, a time-dependent sensor response under temperature modulation has been developed,^4–19 in addition to the strategy of using sensor arrays.²⁰

We have been developing a gas-sensing system based on the dynamic nonlinear response under a cyclic temperature^21–28 or cyclic diffusion²⁹ to obtain multi-dimensional information in addition to one-dimensional information under a steady state. We have reported that the nonlinear dynamic response of a gas sensor changes characteristically depending on the concentration and chemical structure of gas molecules, and these characteristic responses can be qualitatively reproduced by a numerical simulation based on the kinetics of gas molecules (diffusion, adsorption, and reaction) on the sensor surface.^22,25,27,29

In this study, we demonstrated that the nonlinear responses to gaseous samples changed characteristically depending on the scanning profile of the cyclic temperature, which was regulated by the second-harmonic heater voltage, to improve the discrimination of the nonlinear dynamic response. The principle of this system is that the phase of the second-harmonic heater voltage can partially change the dynamic sensor response which depends on the kinetics of gases on the sensor surface. The characteristic nonlinear responses of a semiconductor gas sensor under the application of temperature perturbation including the second harmonic were not only evaluated by the higher harmonics of FFT but also qualitatively reproduced based on a diffusion–reaction model for the solid surface.

2. Experimental

Fig. 1 shows a schematic representation of the experimental apparatus developed from our previous system. The processes (1)–(3) were performed.^22,25,28 Here, (3) is the process developed in this paper. (1) A fundamental harmonic (frequency: f₀ = 0.04 Hz) voltage, V₁, was generated with a waveform generator (NF Electronic Instruments, WF1946, Japan) and supplied to the heater of a semiconductor sensor. Under this condition, the time-variation of temperature was almost sinusoidal, as shown in Fig. 2b.²⁵ (2) The time-developed output signal as the sensor response was characteristically deformed from the sinusoidal input signal of temperature depending on the sample gases, i.e., the sensor response included the higher harmonics which corresponded to the nonlinearity. (3) To increase the discrimination of gases based on the dynamic nonlinear response, the second-harmonic (2f₀ = 0.08 Hz) voltage, V₂, was applied further, i.e., a modulated signal (V_m), which was composed of the sum of the fundamental and second-harmonic voltages (V_m = V₁ + V₂), was supplied to the heater of the gas sensor. In this study, we examined three heater voltages, i.e., V_α = 3.60 + 1.35cos2πf₀t + 0.66cos(4πf₀t + π/6) (V), V_β = 3.91 + 1.68cos2πf₀t (V), and V_γ = 3.52 + 1.43cos2πf₀t (V) + 0.70cos(4πf₀t + 5π/3) (V). The region of temperature operation was constant at 400–560 K for the heater voltage, and to maintain the same temperature region, the amplitude of V₁ was adjusted to be almost twice as large as that of V₂ for an individual θ. To characterize the dynamic sensor responses to gases, the time-variation of the sensor conductance, for which the sample number was 1024 and the wavenumber of the fundamental harmonic was two, was Fourier-transformed to the frequency domain.

Fig. 1 Diagram of the experimental apparatus for measuring the nonlinear dynamic response (conductance ∝ 1/R_S) of a semiconductor gas sensor under the different profiles of a cyclic temperature. A modulated voltage composed of the sinusoidal voltage (f = 0.04 Hz) on channel 1 (1 ch) and second-harmonic voltage (f = 0.08 Hz) on channel 2 (2 ch) was applied to the heater of the sensor (R_H) to increase the discrimination of sample gases.

Fig. 2 Experimental results for (1) the time-variation of the sensor temperature, and (2) the real and imaginary parts of FFT under the application of heater voltage (a) V_α, (b) V_β, and (c) V_γ.

Fig. 2 shows (1) the time-variation of temperature and (2) the real and imaginary parts of FFT under application of the heater voltage ((a) V_α, (b) V_β, and (c) V_γ). The initial value for FFT was the conductance when only R₁ was obtained for FFT of the time-variation of temperature, i.e., I₁ = 0 (Fig. 2). Here, R_n (n = 0, 1, 2, 3, and 4) and I_n (n = 1, 2, 3, and 4) are the amplitudes of the real (cosine) and imaginary (sine) parts of the nth harmonic in FFT, respectively (R₀: dc value of the conductance, R₁ and I₁: amplitudes of the fundamental harmonic (0.04 Hz)).

The temperature-profile was characterized by a rapid increase in temperature at dT/dt > 0 and a slow decrease in temperature at dT/dt < 0 for V_α, an equivalent increase and decrease in temperature for V_β, and a slow increase in temperature at dT/dt > 0 and a rapid decrease in temperature at dT/dt < 0 for V_γ. Thus, the scanning rate of temperature at dT/dt > 0 is V_α > V_β > V_γ, and that at dT/dt < 0 is V_γ > V_β > V_α.

A SnO₂ gas sensor was used as the semiconductor gas sensor (TGS2620, Figaro Engineering Inc., Osaka, Japan).^1,3,30–34 The conductance of the gas sensor was measured with a digital multimeter (Yokogawa, Model 7552, Japan). The ac voltage and the conductance of the gas sensor were successively stored in a personal computer via a GPIB interface. The desired amount of gas was introduced into a glass cell (2 l) via a glass injector. Ambient air was purified through columns composed of silica gel, activated carbon, and CaCl₂ to remove excess water vapor and other impurities. Purified air was used as a control. The surface temperature of the gas sensor was measured with an infrared thermometer (Keyence, IT2-01, Japan).

The fundamental frequency of the heater voltage was chosen as follows. The higher fundamental frequency is much better to rapidly detect a sample gas, but the amplitude of the scanning temperature is decreased with the increase in the frequency on the present system. On the other hand, the larger amplitude of the scanning temperature generally enhances the selectivity of sample gases on the sensor response. However, the frequency should be decreased to obtain the larger amplitude. Therefore, we chose the fundamental frequency at 0.04 Hz based on the above reasons.

3. Results and discussion

a. Experimental results for dynamic sensor responses depending on the scanning rate of the temperature

Fig. 3 shows the experimental results for the sensor response of temperature T (K) versus conductance G (mS) for (a) benzene, (b) toluene, (c) 1-propanol, and (d) propane under the application of V_α, V_β, and V_γ. The nature of the TversusG curve changed characteristically depending on the composition of the gas species and the different responses were further characterized by the addition of the second-harmonic perturbation. The temperature at maximum conductance, T_max, decreased with an increase in the scanning rate of the temperature at dT/dt > 0. Especially, T_max for 1-propanol was markedly shifted in comparison with those of other gases, i.e., T_max changed from 505 K for V_γ to 420 K for V_β. With an increase in the scanning rate of the temperature, the sensor conductance decreased at dT/dt < 0 except for propane, but increased at dT/dt > 0 and T < T_max. Therefore, for benzene, toluene, and 1-propanol, the magnitudes of hysteresis (the difference in TversusG curve between an increase and decrease in the temperature scan) at the higher temperature were higher than those at the lower temperature under the application of V_γ, while opposite results were noted under the application of V_α, as seen in Fig. 3a, b, c. On the other hand, the magnitude of the hysteresis for propane was V_α > V_β > Vγ, as seen in Fig. 3d.

Fig. 3 Experimental results on the sensor temperature T (K) versus sensor conductance G (mS) under the application of heater voltage, V_α, V_β, and V_γ. The sample gases were (a) benzene, (b) toluene, (c) 1-propanol, and (d) propane, and their concentrations were 1000 ppm. All of the TversusG curves rotate clockwise, as indicated by the arrows, drawn with a solid line for temperature scanning at dT/dt > 0, and by a dotted line for that at dT/dt < 0.

b. Evaluation of the dynamic nonlinear response by FFT

To characterize the dynamic sensor responses to gases, the time-variation of G was Fourier-transformed to the frequency domain. The higher harmonics of FFT changed characteristically depending on not only the sample gases but also the profile of the temperature scan, as shown in Fig. 4. The dependence of T_max on the scanning rate of the temperature at dT/dt > 0 is reflected in R₁ and R₃. For example, R₁ and R₃ for 1-propanol under the condition of V_α are similar to those under the condition of V_β, but are clearly different from those under the condition of V_γ. For propane, the magnitude of hysteresis depending on the temperature scan was reflected in I₁, and the characteristic nature of TversusG curve at dT/dt > 0 and V_γ was especially reflected in R₂. The common features of the aromatic vapors (benzene and toluene) were reflected in the imaginary part. On the other hand, the difference in the response between benzene and toluene was reflected in the real part. The local changes in the sensor responses to toluene and 1-propanol near 430 K at dT/dt < 0 and V_α in Fig. 3b and c were reflected in R₄.

Fig. 4 Amplitudes of FFT for the sensor response G for 1000 ppm (A) benzene, (B) toluene, (C) 1-propanol, and (D) propane under the application of heater voltage, V_α (white bar), V_β (gray bar), and V_γ (black bar). The data correspond to those in Fig. 3.

Fig. 5 shows the concentration dependence of R₂ for the sensor conductance for propane under the application of V_α, V_β, and V_γ. The concentration dependence of the higher harmonic characteristically changed under the application of the second harmonic modulation. These results suggest that the sample gases are quantified based on the characteristic concentration dependence of the higher harmonics. Thus, the dynamic nonlinear response can be further characterized by the different temperature scanning with second-harmonic modulation, V_α, V_β, and V_γ.

Fig. 5 Concentration dependence of R₂ for the sensor conductance for propane under the application of V_α (circle), V_β (triangle), and V_γ (square).

The relationship between the sensor response and FFT has been reported previously.^22,25,26 The real part is related to the sensor response at the quasi-steady state, which depends on the reaction kinetics. The imaginary part is related to the imbalance between the diffusion and reaction of sample gases on the sensor surface.

The decrease in T_max with an increase in the scanning rate of the temperature at dT/dt > 0 suggests that the reaction rate is significantly higher than the rate of the temperature scan; i.e., the system reaches diffusion-limited access. T_max is related to the oxidation in the subphase, which is the region heated by the radiant heat from the sensor element. Above T_max, G may decrease since the concentration of the sample gas supplied to the sensor surface is lowered by oxidation at the subphase. The change in T_max for 1-propanol is largest among these gases since the activation energy of the oxidation for 1-propanol is lowest among these gases.

The decrease in G with an increase in the scanning rate of T at dT/dt < 0 and the increase in G with an increase in the scanning rate at dT/dt > 0 and T < T_max suggest that a rapid scan of T moves away from G in the steady state, which minimizes the hysteresis in the GversusT curve. In contrast, the poor sensitivity to the temperature scan between V_α and V_β for benzene suggests that the oxidative reaction rather than diffusion is the rate-limiting step reaction, as seen in Fig. 3a. Thus, the characteristic nature of hysteresis depending on the profile of the temperature scan is caused by the diffusion–reaction kinetics of the sample gas and oxygen on the sensor surface.

To enhance the reproducibility of the sensor response, we performed the following tests. First, we adjusted the actual heater voltage for every sensor to operate the same temperature range. Second, we checked the dynamic sensor responses to propane and ethanol as standard samples, and chose the much better sensor. In this stage, the feature of Gvs.T curve was qualitatively reproduced when a sensor was replaced. The difference in the absolute value of the sensor conductance was minimized by the normalization of the higher harmonics, e.g., (R_n(gas) − R₀(air))/R₀(air).

c. Theoretical simulation of the sensor response under the application of different profiles of cyclic temperature

To clarify the dynamic sensor responses under the application of different profiles of cyclic temperature, a theoretical simulation of the sensor response was performed based on the following model for a semiconductor gas sensor:^{1,25,26,30–34}

O₂(bulk) → 2O(sub) (1)

where S denotes a surface adsorption site, e⁻ is a free electron, O_ad^m− (m = 1 or 2) is an ionosorbed oxygen, O(sub) is an oxygen gas atom activated by sensor heating, g_x is a sample gas x (x = A or B) in the bulk phase (g_x(bulk)) or subphase (g_x(sub)), g_xO_ad^m− is g_x(sub) adsorbed on the oxidized sensor surface, D is the diffusion coefficient for oxygen (D_O2) or sample gas (D_gx), and k_n (n = −1, 1, 2, 3, or 4) is the rate constant.

The kinetics of the reaction in eqn (1)–(4) for sample gases may be expressed by eqn (5)–(8):

d[O_ad^m−]/dt = k₁(S₀ − [O_ad^m−] − [g_xO_ad^m−])[O(sub)] − k₋₁[O_ad^m−] − k₂[O_ad^m−][g_x(sub)] (5)

d[g_xO_ad^m−]/dt = k₂[O_ad^m−][g_x(sub)] − k₃[g_xO_ad^m−] (6)

d[g_x(sub)]/dt = D_gx(P_x − [g_x(sub)]) − k₄[g_x(sub)][O(sub)] − k₂[O_ad^m−][g_x(sub)] (7)

d[O(sub)]/dt = D_O2(P_O2^1/2 − [O(sub)]) + k₋₁[O_ad^m−] − k₁(S₀ − [O_ad^m−] − [g_xO_ad^m−])[O(sub)] − k₄[g_x(sub)][O(sub)] (8)

where [O_ad^m−] denotes the surface concentration of O_ad^m−, [O(sub)] is the concentration of oxygen atoms thermally activated in the subphase, [g_xO_ad^m−] is the surface concentration of g_xO_ad^m−, [g_x(sub)] is the concentration of gas x in the subphase, S₀ is the maximum occupied concentration of adsorbed gas species, P_O2 is the concentration of oxygen in the cell, and P_x is the concentration of gas x in the cell.

The conductance of a semiconductor gas sensor G(T), depends on the surface Schottky barrier height V_s and temperature, since the conductance of a SnO₂ semiconductor gas sensor may be considered a model of barrier-limited conductance:^1,25,30–34

G(T) = G₀exp(−eV_s/kT) (9)

where G₀ denotes the preexponential factor and k is Boltzmann's constant. The surface Schottky height V_s is given by the surface concentration of ionosorbed oxygen N_t:

V_s = eN_t²/2ε_sκ₀N_d (10)

where ε_sκ₀ denotes the semiconductor permittivity, and N_d is the volumetric density of electron density.

If we assume that N_d is significantly larger than N_t, eqn (11) can be obtained from eqn (9) and (10):

G(T) = G₀exp(−V_slN_t²/T) (11)

where V_sl = e²/2ε_sκ₀N_dk and N_t = [O_ad^m−] + [g_xO_ad^m−].

The rate constants depend on the temperature according to the Arrhenius equation in eqn (12):

k_n = F_nexp(−E_n/RT) (12)

where F_n denotes the preexponential factor and E_n is the activation energy (n = 1, −1, 2, 3, or 4).

Fig. 6 shows a closed line for G(T), which is calculated with eqn (5)–(8), (11), and (12), when T is periodically changed as T_α, T_β, and T_γ. Here, T_α, T_β, and T_γ were experimentally reproduced by the fundamental and second harmonics to conform to the experimental results regarding the time-variation of temperature for sensor heating, V_α, V_β, and V_γ, respectively (see Fig. 2), i.e., T_α = 461 + 70cos2πf₀t + 24cos(4πf₀t + 0.27π) (K), T_β = 480 + 80cos2πf₀t (K), and T_γ = 461 + 75cos2πf₀t + 23cos(4πf₀t + 1.87π) (K).

Fig. 6 Numerical simulation of temperature T (K) versus sensor conductance, G (S), curves based on eqn (5)–(8), (11), and (12) when T_α, T_β, and T_γ are applied. The TversusG curves rotate clockwise, as indicated by the arrows in each figure. f₀ = 0.04 Hz, P_O2 = 210 000 ppm, P_x (x = A or B) = 1000 ppm, F₁ = 0.2, E₁ = 1 kJ mol⁻¹, F₋₁ = 600, E₋₁ = 10 kJ mol⁻¹, D_O2 = 2, V_sl = 2500, S₀ = 1, R = 8.314, and G₀ = 10. (a) Gas A: F₂ = 0.5, E₂ = 1 kJ mol⁻¹, F₃ = 1 × 10⁴, E₃ = 2 kJ mol⁻¹, F₄ = 1000, E₄ = 50 kJ mol⁻¹, D_gA = 0.55. (b) Gas B: F₂ = 0.3, E₂ = 0.1 kJ mol⁻¹, F₃ = 350, E₃ = 5 kJ mol⁻¹, F₄ = 3000, E₄ = 55 kJ mol⁻¹, D_gB = 0.5.

In this simulation, the parameters of the reaction kinetics were basically determined by reference to related papers,^25,26,33,34 and were finally adjusted by the approximate calculation in comparison with the experimental results. The dynamic sensor responses were characterized further by the second-harmonic temperature perturbation, i.e., the GversusT curve for gas A changed differently from that for gas B among T_α, T_β, and T_γ. For example, G for gas A is locally increased at 430 K, while that for gas B is not sensitive under dT/dt < 0 for the application of T_α. As shown in Fig. 7, these characteristic enhanced responses with harmonic perturbation are especially reflected in the higher harmonics, e.g., R₂ for gas A were negative and almost zero under T_α and T_γ, but were positive and highly positive for gas B, while I₂ was positive for gas A under T_γ but zero for gas B. Thus, the characteristic responses for gases A and B are similar to those for toluene and propane.

Fig. 7 Numerical results of the amplitude of the FFT for the sensor response G for gases A and B under the application of a cyclic temperature, T_α (white bar), T_β (gray bar), and T_γ (black bar). The data correspond to those in Fig. 5. The vertical values are R_n or I_n for G.

Conclusion

In this study, we have demonstrated that the dynamic sensor responses to sample gases were changed characteristically by the second-harmonic temperature perturbation, T_α and T_γ. These effects are reflected by the higher harmonics of FFT. The characteristic dynamic sensor responses stimulated by the second-harmonic perturbations were calculated based on a reaction–diffusion model for the semiconductor sensor surface. We also demonstrated that the nonlinear sensor response enhanced by temperature perturbation can provide useful information for the analysis of a sample gas, i.e., it will be possible to distinguish gases under the application of V_α and V_γ even if the sensor response to the sample gas is similar to that to another gas under the application of V_β.

Although we have demonstrated only the second-harmonic perturbation in addition to the fundamental harmonic, the selectivity of the sensor response should be enhanced by choosing other higher harmonic perturbations because the nonlinear response changes depending on the chemical species.^2–28 To identify the optimal condition from a theoretical point of view, the effects on the sensor responses under the temperature perturbation should be clarified by further investigation, e.g., the reaction products and their processes on the sensor surface should be simultaneously identified by further analysis, e.g., GC/MS, to confirm the present model.³⁵

In taste and olfaction, living organisms can detect and quantify chemical stimuli even in complex states, such as natural odors in the environment, without chromatographic separation.^2,36,37 In living organisms, information regarding the chemical structure and concentration is transformed into nervous impulses. Since excitation and pulse generation in biomembranes are typical “nonlinear phenomena”, we believe that the utilization of a nonlinear response which corresponds to the higher harmonics in FFT is worth considering to develop a novel chemical sensor which mimics taste and olfaction.³⁷ In addition, gas chromatography is unable to evaluate the effect of competition among gaseous mixtures due to the analytical principle involved, even though gas chromatography is widely used due to its supposed high sensitivity and selectivity. Therefore, we believe that the present system, which considers not only the time-dependent nonlinear response to individual gases but also the effects of the coexistence of mixed gases, may more closely mimic taste and olfaction.

Acknowledgements

We thank Mr T. Ogawara (Kyoto University, Japan) and Dr K. Kato (Figaro Engineering Inc., Japan) for their useful technical suggestions. This study was supported in part by a Grant-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology of Japan (No. 16550124).

References

1. A. Mandelis and C. Christofide, Chem. Anal. (New York), 1993, 125, 19.
2. Sensors and Sensory Systems for an Electronic Nose, ed. J. W. Gardner and P. N. Barlett, NATO ASI Series, Kluwer, Dordrecht, The Netherlands, 1992, p. 212.
3. K. Ihokura and J. Watson, The Stannic Oxide Gas Sensors, Principle and Applications, CRC Press, Boca Raton, 1994.
4. T. Otagawa and J. R. Stetter, Sens. Actuators, 1987, 11, 251.
5. S. Vaihinger, W. Göpel and J. R. Stetter, Sens. Actuators, B, 1991, 4, 337.
6. J. W. Gardner, Sens. Actuators, B, 1990, 1, 166.
7. W. M. Sears and K. Colbow, Sens. Actuators, 1989, 19, 333.
8. S. Wlodek, K. Colbow and F. Consadori, Sens. Actuators, B, 1991, 3, 63.
9. M. Roth, R. Hartinger, R. Faul and H.-E. Endres, Sens. Actuators, B, 1996, 35–36, 358.
10. B. Yea, T. Osaki, K. Sugahara and R. Konishi, Sens. Actuators, B, 1997, 41, 121.
11. A. Heilig, N. Bârsan, U. Weimar, M. Schweizer-Berberich, J. W. Gardner and W. Göpel, Sens. Actuators, B, 1997, 43, 45.
12. K. Yoshikawa, Y. Kato and M. Kitora, Sens. Actuators, B, 1997, 40, 33.
13. T. A. Kunt, T. J. McAvoy, R. E. Cavicchi and S. Semancik, Sens. Actuators, B, 1998, 53, 24–43.
14. A. P. Lee and B. J. Reedy, Sens. Actuators, B, 1999, 60, 35.
15. H. Kohler, J. Röber, N. Link and I. Bouzid, Sens. Actuators, B, 1999, 61, 163.
16. M. Jaelge, J. Wöllenstein, T. Meisinger, H. Böttner, G. Müller, T. Becker and C. B.-v.Braunmühl, Sens. Actuators, B, 1999, 57, 130.
17. K. Kato, Y. Kato, K. Takamatsu, T. Udaka, T. Nakahara, Y. Matsuura and K. Yoshikawa, Sens. Actuators, B, 2000, 71, 192.
18. R. Ionescu and E. Llobet, Sens. Actuators, B, 2002, 81, 289.
19. X. Huang, F. Meng, Z. Pi, W. Xu and J. Liu, Sens. Actuators, B, 2004, 99, 444.
20. K. J. Albert, N. S. Lewis, C. L. Schauer, G. A. Sotzing, S.E. Sitzel, T. P. Vaid and D. R. Walt, Chem. Rev., 2000, 100, 2595.
21. G. J. Gouws and D. J. Gouws, Sens. Actuators, B, 2003, 91, 326.
22. Chemical Analysis Based on Nonlinearity, ed. S. Nakata, NOVA, New York, 2003, ISBN: 1-59033-737-9.
23. S. Nakata, Y. Kaneda, H. Nakamura and K. Yoshikawa, Chem. Lett., 1991, 1505.
24. S. Nakata, H. Nakamura and K. Yoshikawa, Sens. Actuators, B, 1992, 8, 187.
25. S. Nakata, S. Akakabe, M. Nakasuji and K. Yoshikawa, Anal. Chem., 1996, 68, 2067.
26. S. Nakata, E. Ozaki and N. Ojima, Anal. Chim. Acta, 1998, 361, 93.
27. S. Nakata, T. Hashimoto and H. Okunishi, Analyst, 2002, 127, 1642.
28. S. Nakata, H. Okunishi and S. Inooka, Anal. Chim. Acta, 2004, 517, 153.
29. S. Nakata, T. Nakamura, K. Kato, Y. Kato and K. Yoshikawa, Analyst, 2000, 125, 517.
30. N. Yamazoe, Y. Kurokawa and T. Seiyama, Sens. Actuators, 1983, 4, 283.
31. S. Yasunaga, S. Sunahara and K. Ihokura, Sens. Actuators, 1986, 9, 133.
32. Y. Matsuura, K. Takahata and K. Ihokura, Sens. Actuators, 1988, 14, 223.
33. P. K. Clifford and D. T. Tuma, Sens. Actuators, 1982/83, 3, 233.
34. M. J. Madou and S. R. Morrison, Chemical Sensing with Solid State Devices, Academic, New York, 1989.
35. S. Nakata, H. Okunishi, Y. Nakashima and T. Mori, Chem. Lett., 2005, 34, 186.
36. M. Stopfer and G. Laurent, Nature, 1999, 402, 664.
37. P. Èrdi, I. Aradi, Y. Kato and K. Yoshikawa, BioSystems, 1998, 46, 107.

---